AURisk domain
AI use
AI use covers how people use AI tools: unsanctioned tools, personal accounts on AI platforms, sensitive data in prompts and output acted on without validation. The domain treats AI as a data-handling surface and as a judgment surface at once: what goes in can leave the organization, and what comes out can be wrong. Vigilant use counts: using the sanctioned tools and checking AI output against a trusted source. Threat includes AI-generated lures aimed at the person.
Classification
- Dimensions Every signal answers one of 3 questions.
- Maturity practices One capability a program performs, scored on its own: See, Understand, Act, Prove, Govern and Culture.
- Leads to The kind of breach the evidence beneath it points toward, each anchored to a Verizon DBIR or VERIS pattern or an FBI IC3 category.
- NIST CSF 2.0
-
- GV.PO-01
- PR.DS-10
- GV.RM-05
- PR.AT-01 · NIST AI RMF Govern
- Map
- Measure
- MITRE ATT&CK
-
- TA0009 Collection, TA0010 Exfiltration · T1567 · ATLAS AML.T0051 LLM Prompt Injection
Generative AI puts sensitive data and decisions into tools the organization did not choose. A prompt is a data transfer and unvalidated output is an error class of its own, so both need the same measurement as any other data-handling behavior.
Three dimensions in ai use.
- Threat1 signals
- Threat is AI aimed at the person: deepfake or AI-generated lures detected in the channels that reach them.
- Access1 signals
- Access is who administers an enterprise AI platform and can change what every user may send and receive.
- Behavior6 signals
- Behavior is what a person puts into AI and what they do with the result: unauthorized tools, personal accounts, sensitive data in prompts and unvalidated output on the risky side; sanctioned tools and validated output on the vigilant side.
Signals
8 signalsScroll sideways to see every column.
| ID | Signal | Dimension | Polarity | Source |
|---|---|---|---|---|
| AU-T-01 | Targeted by deepfake or AI-generated lures | Threat | Neutral | |
| AU-A-01 | Administers an enterprise AI platform | Access | Neutral | IDP |
| AU-B-01 | Uses unauthorized AI tools | Behavior | Risky | SSE |
| AU-B-02 | Puts sensitive data in AI prompts | Behavior | Risky | SSE |
| AU-B-03 | Uses personal accounts on AI platforms | Behavior | Risky | SSE |
| AU-B-04 | Uses the sanctioned AI tools | Behavior | Vigilant | SSE |
| AU-B-05 | Acts on AI output without validation | Behavior | Risky | SURVEY |
| AU-B-06 | Validates AI output against trusted sources | Behavior | Vigilant | SURVEY |
Headline metrics and grade bands
3 metricsA to F bands for the numbers a program reports in this domain. Bands marked "framework benchmark" are authored targets until the HRM Benchmark publishes measured values.
Scroll sideways to see every band.
| Metric | Unit | A | B | C | D | Anchor |
|---|---|---|---|---|---|---|
| M-AU-1Shadow AI share | % of AI use through personal or unsanctioned accounts | ≤ 10 | ≤ 25 | ≤ 45 | ≤ 67 | DBIR 2026: 67% of AI use ran through non-corporate accounts |
| M-AU-2Sensitive data in prompts | incidents per 100 people per quarter | ≤ 0.2 | ≤ 1 | ≤ 3 | ≤ 6 | Framework benchmark |
| M-AU-3AI output validated before use | % of people who check AI output against a trusted source | ≥ 90 | ≥ 75 | ≥ 60 | ≥ 40 | Framework benchmark; NIST AI RMF Govern |
Indicator library
31 indicators- I.0103Copies enterprise data into personal AI chat tools
- I.0155Processes source code with unapproved AI services
- I.0164Visits websites categorized as AI tools
- I.0437Inputs secrets into AI prompts
- I.0438Submits enterprise data to unmanaged AI platforms
- I.0439Inputs proprietary code into AI tools
- I.0440Inputs customer PII into AI tools
- I.0441Inputs sensitive HR or legal content into AI tools
- I.0442Exposes sensitive prompts during live sessions
- I.0443Uses unapproved AI applications
- I.0444Uses personal AI accounts for work
- I.0445Authorizes AI use outside job scope
- I.0446Implements AI-generated code without validation
- I.0447Sends unverified AI responses to customers
- I.0448Uses AI to craft manipulative or misleading content
- I.0449Inputs third-party copyrighted content into AI
- I.0450Attempts to bypass AI safety controls
- I.0451Employs AI jailbreak techniques
- I.0452Probes AI models to extract training data
- I.0453Persists with unsafe AI prompting
- I.0454Deploys unvetted open-source AI models
- I.0455Disregards enterprise AI policies
- I.0456Accepts AI outputs without validation
- I.0457Reviews prompts for sensitive data
- I.0458Uses enterprise-approved AI platforms
- I.0459Validates AI outputs against trusted sources
- I.0460Discloses AI use in deliverables
- I.0461Documents AI-assisted decisions
- I.0462Avoids AI-generated content in compliance-sensitive contexts
- I.0463Uses pre-approved AI prompt templates
- I.0464Participates in AI model risk assessments
Cohorts in this domain
1Sequences that start here
1"AI use" (AU), Human Risk Management Framework v2.0.0, Living Security, CC BY 4.0. https://www.humanriskmanagement.com/taxonomy/domains/ai-use
Part of the open HRM Taxonomy, CC BY 4.0. Download the taxonomy as JSON · What's new in 2.0.
The initial assessment
Find your level.Then fix what matters.
18 questions, about 4 minutes and nothing to type before the result: your level, a score for each practice and three first playbooks, each Easy, Medium or Hard.